Hi all!
I''m going to set up an active/active OST failover, but gets a litle
confused. In the manual in section 6.4.2 it says:
"To expand the simple two-node example, we add ost2 which is primary on
nodeB, and is on the LUNs nodeB:/dev/sdc1 and nodeA:/dev/sdd1. This is
to demonstrate the /dev/ identify can differ between nodes, but both
devices must map to the same physical LUN." Is this the same disc in a
SAN?
I want to have a local /dev/sdaX on each OST - with same size. That
means:
ost1 with its own /dev/sda3
in an active/active failover with
ost2 with its own /dev/sda3
Is this possible, or have I misunderstanded something?
--
Einar Gautun gauhan@statkart.no
Statens kartverk | Norwegian Mapping Authority
3507 H?nefoss | NO-3507 H?nefoss, Norway
Ph +47 32118372 Fax +47 32118101 Mob +47 92692662
On Mon, 2007-23-04 at 12:56 +0200, Hans Einar Gautun wrote:> I want to have a local /dev/sdaX on each OST - with same size. That > means: > > ost1 with its own /dev/sda3 > in an active/active failover with > ost2 with its own /dev/sda3 > > Is this possible, or have I misunderstanded something?The purpose of failover is so that two OSSes can provide the same service to the lustre network. That is, should one of the two nodes fail (i.e. a piano falls on it and completely destroys it), the other OSS can provide exactly the same services. So you have to ask yourself, if you have a local disk in each machine and one machine gets crushed under a piano, how does other machine provide service from the crushed machine''s disk? There are two solutions. Use a network replication service like drbd (so that the contents of the local disks is identical at all times) or use an external storage system that both OSSes can access, like FC SCSI or Firewire, etc. b.
On Apr 23, 2007 07:05 -0400, Brian J. Murrell wrote:> On Mon, 2007-23-04 at 12:56 +0200, Hans Einar Gautun wrote: > > I want to have a local /dev/sdaX on each OST - with same size. That > > means: > > > > ost1 with its own /dev/sda3 > > in an active/active failover with > > ost2 with its own /dev/sda3 > > > > Is this possible, or have I misunderstanded something? > > The purpose of failover is so that two OSSes can provide the same > service to the lustre network. That is, should one of the two nodes > fail (i.e. a piano falls on it and completely destroys it), the other > OSS can provide exactly the same services. > > So you have to ask yourself, if you have a local disk in each machine > and one machine gets crushed under a piano, how does other machine > provide service from the crushed machine''s disk? > > There are two solutions. Use a network replication service like drbd > (so that the contents of the local disks is identical at all times) or > use an external storage system that both OSSes can access, like FC SCSI > or Firewire, etc.And sometime in the future Lustre will support OST-level RAID, but that is still some ways away. Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc.